Open-source credits

PlayHT is built on open-source software. Thank you to everyone who creates and maintains these projects.

Piper

Licence: MIT. Fast, local neural text to speech, created by Rhasspy and now maintained under the Open Home Foundation. github.com/rhasspy/piper

Piper voices

Licence: Licence varies by voice. Each voice model lists its dataset and licence in its MODEL_CARD file. Check the card for the voice you use before commercial use. huggingface.co/rhasspy/piper-voices

vits-web

Licence: MIT. Runs Piper voices in the browser with ONNX Runtime, by Diffusion Studio. github.com/diffusion-studio/vits-web

espeak-ng (via piper-phonemize)

Licence: GPL-3.0. Turns text into phonemes for Piper. The source code is available from the project. github.com/espeak-ng/espeak-ng

ONNX Runtime Web

Licence: MIT. Runs the neural models in your browser using WebAssembly, by Microsoft. github.com/microsoft/onnxruntime

Kokoro-82M

Licence: Apache 2.0. The neural model behind HD English, by hexgrad. huggingface.co/hexgrad/Kokoro-82M

kokoro-js

Licence: Apache 2.0. Runs Kokoro in the browser. www.npmjs.com/package/kokoro-js

Transformers.js

Licence: Apache 2.0. Machine-learning runtime used by kokoro-js, by Hugging Face. github.com/huggingface/transformers.js

Delivery

The engines are delivered by jsDelivr and the voice files by Hugging Face. Device voices are provided by your operating system or browser and are not part of this site.

Using voices commercially

Kokoro is released under Apache 2.0, which allows commercial use. Piper voices are trained on different datasets, and some carry restrictions such as non-commercial use. Before using a Piper voice commercially, open its MODEL_CARD on Hugging Face and confirm its licence. This page is information, not legal advice.